Figure 3
From: The difficulties of reproducing conventionally derived results through 500k-chip technology
Learning through multiple re-tests. Based on independent training samples (n = 1,042; genotyped for 5,728 SNPs of a conventional 0.4-Mb genome scan) and test samples (n = 746; genotyped for 545,080 SNPs of a 500k-chip), we investigated the extent to which the subjects' individual IgM levels can be reproducibly predicted from a multilocus genotype. To test the reproducibility of predictors across populations and across SNP sets, a "competitive SNP set" approach was used to extract six subsets of 5,728 SNPs each from the 500k-chip data according to the criteria: 1) close vicinity to the original training set; 2) maximum informativeness; and 3) <20% overlap between the six subsets.